67

WhatdoesdecodingfromthePFCrevealabout consciousness? Ned Block 1,* Disputes between rival theories of consciousness have often centered on whether perceptual contents can be decoded from the prefrontal cortex (PFC). Failures to decode from the PFC are taken to challenge ‘cognitive’ theories of consciousness such asthe global workspace theory and higher-order monitoring theories, and decoding successes have been taken to confirm these theories. However, PFC decoding shows both too much and too little. Too much because cognitive theories of consciousness do not need PFC rerepresentation of perceptual contents since pointers to perceptual representations suffice. Too little because there is evidence that PFC decoding of perceptual content reflects postperceptualcognitiverepresentation, such asthoughtsthathavethoseperceptual contents rather than conscious percepts. Decoding from the PFC Recent work on consciousness has emphasized intracranial decoding (see Glossary) of perceptual contents in the PFC. In binocular rivalry, subjects are shown different stimuli in each eye. If the stimuli ‘conflict’ ([1], pp. 191–192) subjects will see first one stimulus, then the other, alternating on and on. Researchers have decoded conscious PFC content in monkeys intracranially (i.e., using electrode arrays penetrating the cortex) during binocular rivalry of perception of up/down or right/left moving grids without any task (using eye movements as a substitute for report). This has been taken to support global workspace and higher-order theories of consciousness [2–4], both of which hold that consciousness requires PFC representation. Confirming the relevance of the decoding to consciousness, decoders that work for binocular rivalry also work for ‘replay’ in which subjects are shown first grids moving one way, then grids moving the other way, in a manner that is supposed to replicate the experience of binocular rivalry. Furthermore, one research group demonstrated that perceptual contents can be decoded from the PFC even whenstimuli are presented quite rapidly in succession, 10 per second [5]. While the authors acknowledge that the stimuli may have been perceived unconsciously, they argue that this finding provides some support for cognitive theories of consciousness since each stimulus masks the previous one, discouraging postperceptual processing, which, as we will see later, provides an alternative explanation for PFC decoding. However, another research group [6] found no difference between conscious and unconscious processing in the PFC, concluding (p. 295), ‘our results challenge the pre-frontal theories in that prefrontal connectivity was not modulated by conscious awareness.’ Again, the success or failure of PFC decoding is taken to be crucial. Templeton World Charity Foundation has funded several ‘adversarial collaborations’ in which advocates of rival theories of consciousness have devised experiments that could challenge each of the rival theories. The first results [7] found PFC decoding for faces and letters but not Highlights A number of recent articles have shown intracranial decoding in prefrontal cortex (PFC) for high-level perceptual features but not low-level features. Furthermore, sustained perception results only in momentary PFC decoding. These results have been taken to both confirm and challenge aspects of ‘prefrontalist’ theories of consciousness. New ‘inattentional blindness’ results have been used to argue for PFC representation of perceptual contents. Pointer versions of prefrontalist theories of consciousness provide a way of accommodating the decoding results but have been thought to be challenged by inattentional blindness results. Pointer theories run into problems with intracranial stimulation. Bifurcation dynamics yields a new noreport paradigm. The global playground provides a better account of pure access than the global workspace. PFC decoding may depend on postperceptual cognition. 1New York University, 5 Washington Place, New York, NY 10003, USA *Correspondence: ned.block@nyu.edu (N. Block). Trends in Cognitive Sciences, Month 2024, Vol. xx, No. xx https://doi.org/10.1016/j.tics.2024.05.004 1 ©2024Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.Trends in Cognitive Sciences for orientations of faces; the former was classified as a win for global workspace theories, and the latter was classified as a challenge for them [7] (even though the face orientations were never task-relevant). As a news report in Science put it, ‘When it came to decoding different categories of objects, the data provided strong support for global neuronal workspace theory (GNWT). But when it came to decoding the orientation of faces, IIT was the better fit’ [5]. (IIT is Integrated Information Theory, an approach that emphasizes posterior visual areas.) Given the fact that both conscious and unconscious contents can be decoded from posterior visual areas even in anesthetized animals [8,9], decoding from the PFC has seemed pivotal to the fortunes of cognitive theories of perceptual consciousness even when it conflicts with reports (Box 1). This article argues that the emphasis on decoding from the PFC is a mistake even for prefrontalist theories. I will argue that there are advantages to prefrontalist theories that do not require decoding from the PFC and further that decoding from the PFC does not show what many of its adherents have supposed. Whycognitive theories of consciousness do not require PFC decoding Let us next consider why advocates of cognitive theories of consciousness have emphasized decoding from the PFC, asking whether decoding from the PFC is truly important to the aims of these theories. Cognitive versus localist theories of consciousness Consciousness research has been dominated by a conflict between two approaches. According to ‘cognitive’ theories of consciousness (notably, global workspace and higher-order thought theories), consciousness of a stimulus is based in processing involving thought, reasoning, noticing, and other forms of cognitive ‘access’ to the perceptible properties of the stimulus based in the PFC; these cognitions can be (but need not be) automatic, involuntary, and implicit. These theorists are sometimes called the ‘prefrontalists.’ (Please note that I use the terms ‘prefrontalist’ and ‘PFC’ to indicate the commitments of cognitive theories of consciousness. However, global workspace theories also emphasize parietal cortex, and, even when it comes to the PFC, what is really at issue is certain regions in front of the central sulcus, notably dorsolateral, ventrolateral, medial prefrontal, anterior cingulate, and orbitofrontal parts of the PFC [3,10–12].) The alternative sensory approach is ‘localist’ in that it holds that perceptual consciousness is based in the circuits that process visual content in posterior sensory areas, although some localists also allow that cognitive phenomenology might be based partly in the PFC [13–15] and that perhaps some high-level perceptual contents [16,17] might be partially localized in the Box 1. When decoding from the PFC conflicts with reports Three different laboratories [7,43,44] found only a few hundred milliseconds of PFC/parietal activation, even though subjects were looking attentively at the stimulus for as long as a second and a half. Although reported consciousness has been the gold standard measure of awareness for prefrontalist approaches, this result has led some researchers to take seriously the idea that when you stare at something attentively for a few seconds, you are only conscious of it for a few hundredmilliseconds, despite viewers’ reports that they are conscious of it the whole time; that is, decoding has been taken to trump reports. In one study [43], the possibility that these transient PFC activations may be taken by some to indicate momentary consciousness despite reports to the contrary is mentioned fivetimes(pp.1,2,9,10,11).They consider the possibility that what seems to be continuous perception of a stimulus is composed of discrete samples, saying (p. 9), ‘…the introspective subjective percept is of stable continuous images with varying durations. Taken at face value, this would suggest that the sustained and stable visual representations in the ventral visual stream underpin our ongoing consciousexperience. However, to the extent that perception is composed of discrete samples, each generating a transient ignition, the frontoparietal response would correspond more directly to experience.’ This line of thought would fit with some theorists’ interpretation of ‘inattentional blindness’ in which what seems to be a constant ongoing perception of details in the environment does not reflect what is really happening in conscious perception [52]. Glossary Change blindness: when subjects fail to report changes in stimuli even when looking at them. Change blindness may be a matter of failing to apply concepts rather than a failure of perception. Decoding: the use of brain imaging to reconstruct how the perceiver is representing the stimulus. Dereferencing: a process by which the representation that is the object of a pointer is retrieved for further processing or use. Global workspace theory of consciousness: holds that a perceptual representation is conscious if, and only if, it is transmitted, via longrange connections between perceptual and cognitive areas of the brain in the prefrontal cortex, so as to be globally available for processing by cognitive mechanisms. Higher-order theories of consciousness: holdthata perception is conscious in virtue of another state that represents the perception. According to traditional ‘rerepresentation’ higher-order theories, for a perceptual representation of a square to beconscious, there must bea thought (usually considered to be based in the prefrontal cortex) about it to the effect that one is seeing a square, thus involving two representations of squareness. Inattentional blindness: when subjects fail to report unusual stimuli if their attention is distracted. Inattentional blindness may be a matter of failing to apply concepts rather than a failure of perception. Information Integration Theory (IIT): differs from global workspace and higher-order accounts ofconsciousness in that it can be seen as characterizing the relation ‘x is more conscious than y’ rather than consciousness per se. IIT says roughly that x is more conscious than y to the extent that x is more differentiated than y and those differentiated elements are more integrated. Postperceptual processing: processing applied to a perceptual representation, often for cognitive purposes. Conceptualization of a perceptual representation is one important type of postperceptual processing. Pointer: a register can contain the address of another register, in which case the first register is regarded as 2 TrendsinCognitive Sciences, Month 2024, Vol. xx, No. xxTrends in Cognitive Sciences PFC. According to localists, decoding of perceptual contents in the PFC normally reflects postperceptual cognitive processing. Some localists focus on what they call the ‘hot zone’ in posterior visual areas because their theories favor the kind of cortical organization found in perceptual areas [18]. The higher-order and global workspace camps of prefrontalists [19] recently joined forces to argue that their cognitive approach is the most promising approach to machine consciousness. No report paradigms Early results on the neural correlates of consciousness favored the PFC as an essential part of the neural basis of consciousness [20]. However, these early studies tended to focus on reported conscious states, thereby conflating the neural basis of consciousness with the decision and evaluation processes required to decide which response to make in order to report consciousness, as I pointed out in a series of articles [21–23]. Likewise, others [24] have pointed out that attentional activations prior to the stimulus also contaminated attempts to find the neural correlates of consciousness. The major idea for dealing with this conflation was ‘no-report’ paradigms [25]. No-report paradigms rely on reports, but only indirectly. In no-report paradigms, reports are used to validate a measure of consciousness in some subjects; then, in other subjects, the measure can be used without the report. In the intracranial studies of binocular rivalry mentioned earlier [2,4], eye movements (optokinetic nystagmus) tell us whether the subject is experiencing motion in one direction or the other and are a good enough indicator of what the subject is experiencing to substitute for report. In some no-report paradigms, there is no task; in others, there are tasks that are not relevant to the variable of interest and do not affect responses in a relevant way. Another kind of no-report paradigm is discussed in the last section. Many studies have found decoding from the PFC for conscious contents using no-report paradigms [2,7,26–28]. Rich versus sparse perceptual representation Another aspect of the prefrontalist/localist controversy is that perceptual theories suggest ‘rich’ conscious perception because many perceptual representations can be active at once, whereas cognitive theories suggest that consciousness is only as rich as our thinking about andcognitively accessing our perceptual states. Our cognitive concepts of our perceptions are much sparser than the perceptual contents themselves. For example, there are thousands of discriminable sounds but many fewer concepts of them [29]. Even people with perfect pitch can recognize only around 100 pitches [30]. Prefrontalists argue that psychological phenomena suchasinattentionalblindness and change blindness support the view that consciousness is sparse. Researchers have shown subjects naturalistic pictures and videos that change in various ways outside the fixation point, that point being specified by eye trackers. The pictures and videos were normal at the fixation point but were deprived of color or normal form outside the fixation point (the latter using texturization algorithms [31]). The overall result has been that subjects often fail to notice the weirdness of the peripheral displays [32,33]. Because in these paradigms the PFC representations are assumed to be ‘normal’ but perceptual representations of the environment in posterior visual areas are assumed to be abnormal (fitting the actual texturized or colorless periphery of the perception) but do not result in noticing, these results have been taken to support cognitive theories of consciousness. (I have argued against such ideas [1], but I will not discuss my objections here.) Pointer theories This article concerns two major reasons to think this emphasis on decoding from the PFC is a mistake. First, as recent versions of the higher-order approach have emphasized, consciousness